Re: RFA: loop optimization bug

[Andrew Pinski <pinskia at physics dot uc dot edu>]

On Jul 27, 2005, at 1:09 PM, Dale Johannesen wrote:

> On Jul 27, 2005, at 1:58 AM, Zdenek Dvorak wrote:
> > tree optimizers in general do not care about order of blocks.  I have
> > some patch to change this (more or less only to improve readability of
> > dumps), I don't know whether it would help this particular case or
> > not.
>
> I actually don't think the tree opts should care.  You've got a CFG and
> all these spiffy graph-based algorithms, why not use them?   The 
> problem
> as I see it is that the (old) RTL loop opt does care.  If that's 
> unavoidable,
> IMO a better approach would be to force the blocks into some canonical
> order somewhere between RTL conversion and the loop optimizer.

Or even better just remove the old RTL loop opt.

-- Pinski